Profiling the Fuzzy Latent Structure of Multidimensional Poverty: Toward Valuable Insights for Poverty Policymakers

Abstract

Several multidimensional poverty indices have been proposed, and have been extensively studied in the literature. On the other hand, the need for aggregation of poverty indicators into one multidimensional index has been questioned. It has been argued even so that this aggregation can be misleading for political targeting strategies. Subsequently, some researchers have advocated that the use of the latent class analysis would address these issues. However, this setting does not allow to take into account the fuzzy nature of the latent poverty concept. The contribution here is to use the Grade-of-Membership (GoM) model to profile the fuzzy latent structure of multidimensional poverty, for a more realistic handling of this phenomenon. The application of the GoM methodology to multivariate poverty data for the Tunisian case reveals four most prevalent multidimensional poverty profiles. The results emphasize the role played by contextual effects. Indeed, the rural cluster is suffering more intense deprivation and groups in the central and coastal regions have a more comfortable status in comparison with the group of households residing in inland regions. A thorough analysis of these patterns is put forward in this research, giving valuable insights to policy makers.

Keywords:

• multidimensional poverty analysis
• fuzzy sets theory
• Grade of Membership models
• poverty policy strategies

JEL Classification Codes::

• I32

Notes

1 The software chosen to run this model is the package mixedMem in the R statistical software, developed in 2015 by Y. Samuel Wang and Elena A. Erosheva from the University of Washington. In this package, the variational EM algorithm is used until convergence to derive the estimated parameters of the model. More details about the variational algorithm and other estimation techniques and algorithms used for estimation of GoM models can be found in the works of Matthew Beal (2003); David Blei, Andrew Y Ng, and Michael I. Jordan (2003); Edoardo Airoldi et al. (2014); and Elena Erosheva, Stephen Fienberg, and Cyrille Joutard (2007).

Additional information

Notes on contributors

Asma Zedini

Asma Zedini is currently an assistant professor of quantitative methods at Avicenne Private Business School, Tunisia.